Professional Documents
Culture Documents
Unit 1: Generating New Images Using Gans: Week 6: Advanced Deep Learning Topics
Unit 1: Generating New Images Using Gans: Week 6: Advanced Deep Learning Topics
Pixel
Noise:
Pixel
Pixel
Pixel
© 2017 SAP SE or an SAP affiliate company. All rights reserved. ǀ PUBLIC 4
Generating New Images Using GANs
Generator
Noise: Fake:
Unflatten
Generator
Classification:
© 2017 SAP SE or an SAP affiliate company. All rights reserved. ǀ PUBLIC Real/Fake? 6
Generating New Images Using GANs
Discriminator
Pixel
Pixel
0/1 Real/Fake?
ö
Pixel
Pixel
Fake or Real
Image
Unflatten
Discriminator Real/Fake?
Fake Image
(from Generator)
Discriminator Fake!
Real Image
(from Dataset)
Discriminator Real!
Noise:
(Frozen:)
Pixel Pixel
0/
1
Pixel Pixel
Pixel Pixel
Pixel
0/1 Real!
Pixel
Pixel
Conditional GANs:
▪ GANs can also be used in a more supervised
setting, where G and D are also fed labels
▪ This allows you to feed the network’s labels at
runtime to generate specific outputs!
open@sap.com
© 2017 SAP SE or an SAP affiliate company. All rights reserved.
No part of this publication may be reproduced or transmitted in any form or for any purpose without the express permission of SAP SE or an SAP affiliate company.
The information contained herein may be changed without prior notice. Some software products marketed by SAP SE and its distributors contain proprietary software components
of other software vendors. National product specifications may vary.
These materials are provided by SAP SE or an SAP affiliate company for informational purposes only, without representation or warranty of any kind, and SAP or its affiliated
companies shall not be liable for errors or omissions with respect to the materials. The only warranties for SAP or SAP affiliate company products and services are those that are
set forth in the express warranty statements accompanying such products and services, if any. Nothing herein should be construed as constituting an additional warranty.
In particular, SAP SE or its affiliated companies have no obligation to pursue any course of business outlined in this document or any related presentation, or to develop or release
any functionality mentioned therein. This document, or any related presentation, and SAP SE’s or its affiliated companies’ strategy and possible future developments, products,
and/or platform directions and functionality are all subject to change and may be changed by SAP SE or its affiliated companies at any time for any reason without notice. The
information in this document is not a commitment, promise, or legal obligation to deliver any material, code, or functionality. All forward-looking statements are subject to various
risks and uncertainties that could cause actual results to differ materially from expectations. Readers are cautioned not to place undue reliance on these forward-looking statements,
and they should not be relied upon in making purchasing decisions.
SAP and other SAP products and services mentioned herein as well as their respective logos are trademarks or registered trademarks of SAP SE (or an SAP affiliate company)
in Germany and other countries. All other product and service names mentioned are the trademarks of their respective companies.
See http://global.sap.com/corporate-en/legal/copyright/index.epx for additional trademark information and notices.
Week 6: Advanced Deep Learning Topics
Unit 2: Reinforcement Learning
Reinforcement Learning
Applications of machine learning
REINFORCEMENT
▪ ~100,000s of examples
sample #1
– Each example consists of
▫ Data [ Vector ]
3
▫ Label(s)
…
.
N
.
.
Environment
© 2017 SAP SE or an SAP affiliate company. All rights reserved. ǀ PUBLIC 5
Reinforcement Learning
Machine Learning – Reinforcement Learning Agent
RL Feedback Loop
▪ At each step the agent
– Executes action: 𝐴𝑡
– Receives observation: 𝑂𝑡
– Receives reward: 𝑅𝑡
▪ The environment
– Receives action: 𝐴𝑡
– Emits observation: 𝑂𝑡+1
– Emits reward: 𝑅𝑡+1
Environment
© 2017 SAP SE or an SAP affiliate company. All rights reserved. ǀ PUBLIC 6
Reinforcement Learning
Applying RL to games
Atari Example
▪ Agent is a DL network
– Interprets screen pixels
– Outputs game action
– Possible to use CNN+RNN
▪ Environment is Atari Emulator
– Game AI
▪ Reward is game score
open@sap.com
© 2017 SAP SE or an SAP affiliate company. All rights reserved.
No part of this publication may be reproduced or transmitted in any form or for any purpose without the express permission of SAP SE or an SAP affiliate company.
The information contained herein may be changed without prior notice. Some software products marketed by SAP SE and its distributors contain proprietary software components
of other software vendors. National product specifications may vary.
These materials are provided by SAP SE or an SAP affiliate company for informational purposes only, without representation or warranty of any kind, and SAP or its affiliated
companies shall not be liable for errors or omissions with respect to the materials. The only warranties for SAP or SAP affiliate company products and services are those that are
set forth in the express warranty statements accompanying such products and services, if any. Nothing herein should be construed as constituting an additional warranty.
In particular, SAP SE or its affiliated companies have no obligation to pursue any course of business outlined in this document or any related presentation, or to develop or release
any functionality mentioned therein. This document, or any related presentation, and SAP SE’s or its affiliated companies’ strategy and possible future developments, products,
and/or platform directions and functionality are all subject to change and may be changed by SAP SE or its affiliated companies at any time for any reason without notice. The
information in this document is not a commitment, promise, or legal obligation to deliver any material, code, or functionality. All forward-looking statements are subject to various
risks and uncertainties that could cause actual results to differ materially from expectations. Readers are cautioned not to place undue reliance on these forward-looking statements,
and they should not be relied upon in making purchasing decisions.
SAP and other SAP products and services mentioned herein as well as their respective logos are trademarks or registered trademarks of SAP SE (or an SAP affiliate company)
in Germany and other countries. All other product and service names mentioned are the trademarks of their respective companies.
See http://global.sap.com/corporate-en/legal/copyright/index.epx for additional trademark information and notices.
Week 6: Advanced Deep Learning Topics
Unit 3: Unsupervised Learning
Unsupervised Learning
Applications of machine learning
REINFORCEMENT
▪ ~100,000s of examples
sample #1
– Each example consists of
▫ Data [ Vector ] 3
▫ Label(s)
…
.
N
.
.
෩
X−X
© 2017 SAP SE or an SAP affiliate company. All rights reserved. ǀ PUBLIC Reconstruction Error 5
Unsupervised Learning
DL anomaly detection in time series
w
1
w
2
w
3
… … w
N
Output (Reconstruction)
Reconstruction error (RE) as a proxy to outliers
▪ Whenever RE is high, it indicates something
– Threshold can be set using statistical bounds
Reconstruction vs Input
Input
© 2017 SAP SE or an SAP affiliate company. All rights reserved. ǀ PUBLIC 7
Unsupervised Learning
Interpreting anomalies
open@sap.com
© 2017 SAP SE or an SAP affiliate company. All rights reserved.
No part of this publication may be reproduced or transmitted in any form or for any purpose without the express permission of SAP SE or an SAP affiliate company.
The information contained herein may be changed without prior notice. Some software products marketed by SAP SE and its distributors contain proprietary software components
of other software vendors. National product specifications may vary.
These materials are provided by SAP SE or an SAP affiliate company for informational purposes only, without representation or warranty of any kind, and SAP or its affiliated
companies shall not be liable for errors or omissions with respect to the materials. The only warranties for SAP or SAP affiliate company products and services are those that are
set forth in the express warranty statements accompanying such products and services, if any. Nothing herein should be construed as constituting an additional warranty.
In particular, SAP SE or its affiliated companies have no obligation to pursue any course of business outlined in this document or any related presentation, or to develop or release
any functionality mentioned therein. This document, or any related presentation, and SAP SE’s or its affiliated companies’ strategy and possible future developments, products,
and/or platform directions and functionality are all subject to change and may be changed by SAP SE or its affiliated companies at any time for any reason without notice. The
information in this document is not a commitment, promise, or legal obligation to deliver any material, code, or functionality. All forward-looking statements are subject to various
risks and uncertainties that could cause actual results to differ materially from expectations. Readers are cautioned not to place undue reliance on these forward-looking statements,
and they should not be relied upon in making purchasing decisions.
SAP and other SAP products and services mentioned herein as well as their respective logos are trademarks or registered trademarks of SAP SE (or an SAP affiliate company)
in Germany and other countries. All other product and service names mentioned are the trademarks of their respective companies.
See http://global.sap.com/corporate-en/legal/copyright/index.epx for additional trademark information and notices.
Week 6: Advanced Deep Learning Topics
Unit 4: Deep Learning on Mobile
Deep Learning on Mobile
Getting to a trained model
– e.g. 64GBs
▪ Battery efficiency is a bottleneck!
– 20-44 hrs depending on mode & consumption
Layer Subsets
– Especially critical for applications that process 2
capacity ]
streaming data
1
Weight Pruning
▪ Opportunistically remove weights
– Can lead to significant parameter reduction
– Requires a [full] model to be initially trained
– Inspired by synaptic pruning in biological neurons
▪ Continue streaming training data
– Monitor performance loss due to reduced weights
▪ Target layers with highest battery impact
– Requires energy estimation model
▫ Track load, multiply, and accumulate calls
[ intermediate representation ]
– e.g. Python code
Compute Graph
▪ Intermediate representation
– Compilation and optimization
– Target-independent
▪ Target-dependent optimizations
– Code generation
[ pruning + quantization ]
▪ Runtime
Optimizations
– Packaged binary +
Runtime
© 2017 SAP SE or an SAP affiliate company. All rights reserved. ǀ PUBLIC 7
Deep Learning on Mobile
Available tools
open@sap.com
© 2017 SAP SE or an SAP affiliate company. All rights reserved.
No part of this publication may be reproduced or transmitted in any form or for any purpose without the express permission of SAP SE or an SAP affiliate company.
The information contained herein may be changed without prior notice. Some software products marketed by SAP SE and its distributors contain proprietary software components
of other software vendors. National product specifications may vary.
These materials are provided by SAP SE or an SAP affiliate company for informational purposes only, without representation or warranty of any kind, and SAP or its affiliated
companies shall not be liable for errors or omissions with respect to the materials. The only warranties for SAP or SAP affiliate company products and services are those that are
set forth in the express warranty statements accompanying such products and services, if any. Nothing herein should be construed as constituting an additional warranty.
In particular, SAP SE or its affiliated companies have no obligation to pursue any course of business outlined in this document or any related presentation, or to develop or release
any functionality mentioned therein. This document, or any related presentation, and SAP SE’s or its affiliated companies’ strategy and possible future developments, products,
and/or platform directions and functionality are all subject to change and may be changed by SAP SE or its affiliated companies at any time for any reason without notice. The
information in this document is not a commitment, promise, or legal obligation to deliver any material, code, or functionality. All forward-looking statements are subject to various
risks and uncertainties that could cause actual results to differ materially from expectations. Readers are cautioned not to place undue reliance on these forward-looking statements,
and they should not be relied upon in making purchasing decisions.
SAP and other SAP products and services mentioned herein as well as their respective logos are trademarks or registered trademarks of SAP SE (or an SAP affiliate company)
in Germany and other countries. All other product and service names mentioned are the trademarks of their respective companies.
See http://global.sap.com/corporate-en/legal/copyright/index.epx for additional trademark information and notices.
Week 6: Advanced Deep Learning Topics
Unit 5: Summary, Recap, and Further Resources
Summary, Recap, and Further Resources
Summary and recap
open@sap.com
© 2017 SAP SE or an SAP affiliate company. All rights reserved.
No part of this publication may be reproduced or transmitted in any form or for any purpose without the express permission of SAP SE or an SAP affiliate company.
The information contained herein may be changed without prior notice. Some software products marketed by SAP SE and its distributors contain proprietary software components
of other software vendors. National product specifications may vary.
These materials are provided by SAP SE or an SAP affiliate company for informational purposes only, without representation or warranty of any kind, and SAP or its affiliated
companies shall not be liable for errors or omissions with respect to the materials. The only warranties for SAP or SAP affiliate company products and services are those that are
set forth in the express warranty statements accompanying such products and services, if any. Nothing herein should be construed as constituting an additional warranty.
In particular, SAP SE or its affiliated companies have no obligation to pursue any course of business outlined in this document or any related presentation, or to develop or release
any functionality mentioned therein. This document, or any related presentation, and SAP SE’s or its affiliated companies’ strategy and possible future developments, products,
and/or platform directions and functionality are all subject to change and may be changed by SAP SE or its affiliated companies at any time for any reason without notice. The
information in this document is not a commitment, promise, or legal obligation to deliver any material, code, or functionality. All forward-looking statements are subject to various
risks and uncertainties that could cause actual results to differ materially from expectations. Readers are cautioned not to place undue reliance on these forward-looking statements,
and they should not be relied upon in making purchasing decisions.
SAP and other SAP products and services mentioned herein as well as their respective logos are trademarks or registered trademarks of SAP SE (or an SAP affiliate company)
in Germany and other countries. All other product and service names mentioned are the trademarks of their respective companies.
See http://global.sap.com/corporate-en/legal/copyright/index.epx for additional trademark information and notices.